vs
13:4
overall capability wins (0 ties across 17 scored). Kimi K3 leads overall.
- Reasoning5:1
- Coding4:0
- Arena1:0
- Agents0:1
- Tool use1:1
- Composite indices1:1
- Pricing0:3
- Speed0:2
- Specs2:1
Verdict
Kimi K3 leads coding (widest gap: +12.4 on DeepSWE); Kimi K3 ranks higher on LMArena (+4 Elo); Kimi K3 offers larger context (1.0M vs 1M).
- Kimi K3 leads coding (widest gap: +12.4 on DeepSWE)
- Kimi K3 ranks higher on LMArena (+4 Elo)
- Kimi K3 offers larger context (1.0M vs 1M)
- Qwen3.8 Max allows ~2.0x more max output (131K vs 64K)
- Qwen3.8 Max is ~2.0x cheaper on a blended token basis than Kimi K3
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Kimi K3
- $6.75
- Qwen3.8 Max
- $3.50
Qwen3.8 Max is about 1.9x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Moonshot
Alibaba
Family
Kimi
Qwen
License
Kimi K3 License
Qwen3.8-Max License
Open weights
Yes
Yes
Release
Jul 16, 2026
Aug 3, 2026
Knowledge cutoff
-
-
API / provider
Moonshot
Alibaba Cloud
Modalities
text, image, video → text
text, image, video → text
Specs
Context window
1.0M
1M
Max output
64K
131K
Parameters
2.8T (104B act.)
2.4T (95B act.)
Pricing
Input $/1M
$3
$2
Output $/1M
$15
$6
Blended $/1M (3∶1)
$6
$3
Speed
tok/s
60
86
TTFT (s)
0.6
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
93.5
92.8
Humanity's Last Exam
46.9
43.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
47.6
31.7
AA-LCR v1.1
88.7
80.3
CritPt
23.4
17.7
MMMU-Pro
80.5
82.8
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
67.7
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
85.0
81.3
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
59.5
52.1
CursorBench
60.8
—
SWE-Rebench
—
—
NL2Repo-Bench
—
55.9
DeepSWE
69.0
56.6
WebDev Arena
1,674
1,671
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,485
1,481
| Metric | Kimi K3 | Qwen3.8 Max | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Moonshot | Alibaba | — |
| Family | Kimi | Qwen | — |
| License | Kimi K3 License | Qwen3.8-Max License | — |
| Open weights | Yes | Yes | — |
| Release | Jul 16, 2026 | Aug 3, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Moonshot | Alibaba Cloud | — |
| Modalities | text, image, video → text | text, image, video → text | — |
| Specs | |||
| Context window | 1.0M | 1M | A +49K |
| Max output | 64K | 131K | B +67K |
| Parameters | 2.8T (104B act.) | 2.4T (95B act.) | A +400 |
| Pricing | |||
| Input $/1M | $3 | $2 | B +$1 |
| Output $/1M | $15 | $6 | B +$9 |
| Blended $/1M (3∶1) | $6 | $3 | B +$3 |
| Speed | |||
| tok/s | 60 | 86 | B +26 |
| TTFT (s) | 0.6 | 0.4 | B +0.20 |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 93.5 | 92.8 | A +0.7 pts |
| Humanity's Last Exam | 46.9 | 43.1 | A +3.8 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 47.6 | 31.7 | A +15.9 pts |
| AA-LCR v1.1 | 88.7 | 80.3 | A +8.4 pts |
| CritPt | 23.4 | 17.7 | A +5.7 pts |
| MMMU-Pro | 80.5 | 82.8 | B +2.3 pts |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | 67.7 | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 85.0 | 81.3 | A +3.7 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 59.5 | 52.1 | A +7.4 pts |
| CursorBench | 60.8 | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | 55.9 | — |
| DeepSWE | 69.0 | 56.6 | A +12.4 pts |
| WebDev Arena | 1,674 | 1,671 | A +4 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,485 | 1,481 | A +4 Elo |
Kimi K3
- Context
- 1.0M / 64K out
- Parameters
- 2.8T (104B act.)
- Price
- $3 / $15
- Speed
- 60 tok/s · 0.6s TTFT
- Modalities
- text, image, video → text
- License
- Kimi K3 License
Largest open-weight model (2.8T MoE / 104B active) — official card reports strong reasoning and agentic coding results.
Qwen3.8 Max
- Context
- 1M / 131K out
- Parameters
- 2.4T (95B act.)
- Price
- $2 / $6
- Speed
- 86 tok/s · 0.4s TTFT
- Modalities
- text, image, video → text
- License
- Qwen3.8-Max License
Alibaba's Aug 2026 Max flagship (2.4T MoE / 95B active) — frontier agentic coding via QwenCloud / DashScope with 1M multimodal context and $2/$6 API pricing. Open weights released Aug 12, 2026 as Qwen3.8-2.4T-A95B (text-only, thinking-mandatory, Qwen3.8-Max license); the hosted version adds vision input, non-thinking mode, and built-in tools.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.